arXiv:1811.09047v1 [cs.DC] 22 Nov 2018 · smarter planet [1, 2]. From the IoT environment, Big Data are generated in each and every moment from sensors, messaging systems, mobile

“Sample”2018/11/26page 1

Chapter 1

Fog Computing Architecture: Survey andChallenges

Ranesh Kumar Naha1 Saurabh Garg 2 andAndrew Chan3

Abstract: Emerging technologies that generate a huge amount of data such as theInternet of Things (IoT) services need latency aware computing platforms to supporttime-critical applications. Due to the on-demand services and scalability features ofcloud computing, Big Data application processing is done in the cloud infrastructure.Managing Big Data applications exclusively in the cloud is not an efficient solutionfor latency-sensitive applications related to smart transportation systems, healthcaresolutions, emergency response systems and content delivery applications. Thus, theFog computing paradigm that allows applications to perform computing operationsin-between the cloud and the end devices has emerged. In Fog architecture, IoT de-vices and sensors are connected to the Fog devices which are located in close prox-imity to the users and it is also responsible for intermediate computation and storage.Most computations will be done on the edge by eliminating full dependencies on thecloud resources. In this chapter, we investigate and survey Fog computing architec-tures which have been proposed over the past few years. Moreover, we study therequirements of IoT applications and platforms, and the limitations faced by cloudsystems when executing IoT applications. Finally, we review current research worksthat particularly focus on Big Data application execution on Fog and address severalopen challenges as well as future research directions.

1.1 Introduction

IoT is a connected network of things where all connected nodes are constantly com-municating together automatically in coordination to produce collective results inorder to serve people for a better life and economic advancement. It promises to

1School of Technology, Environments and Design, University of Tasmania, Hobart, TAS 7001, AustraliaEmail:[email protected] of Technology, Environments and Design, University of Tasmania, Hobart, TAS 7001, AustraliaEmail:[email protected] of Engineering, University of Tasmania, Hobart, TAS 7001, AustraliaEmail:[email protected]

arX

iv:1

811.

0904

7v1

[cs

.DC

] 2

2 N

ov 2

018


2

serve everyone to get access to any service or network related to objects, people,programs and data from anywhere. IoT also enables communication between ma-chines, people and things to implement an environment for smart living, smart cities,smart transportation systems, smart energy distribution, smart health services, smartindustries, smart buildings, and they all work together towards making this planet asmarter planet [1, 2].

From the IoT environment, Big Data are generated in each and every momentfrom sensors, messaging systems, mobile devices and social networks resulting ina new form of network architecture. There are many technical challenges relatedto the IoT environment arising due to its distributed, complex and dynamic nature.These technical challenges include connectivity, capacity, cost, power, scalabilityand reliability [1]. Another key issue is the processing of Big Data that is producedfrom various IoT nodes. Generally, we rely on the cloud to process Big Data butsometimes it is really not feasible to transfer all generated data to the cloud for pro-cessing and storage. The process of sending all generated data to the cloud mightoccupy a certain amount of network bandwidth and on the other hand, the cloud isnot able to process latency aware applications due to high response times. Hence,Fog computing came into the picture to process Big Data near to user locations.

Processing Big Data from the Fog environment is an emerging computing paradigmwhich brings application services closer to the users with better service quality. Anydevice that has converged infrastructure can act as a Fog node and provide computa-tion, storage, and networking services to the users. Fog devices are also connected tothe cloud to handle complex processing and long-term storage, and this computingparadigm also referred as IoT-Fog-cloud framework [3]. Figure 1.1 shows the flowand the processing of Big Data in the IoT environment.

1.2 Fog computing architecture

Fog computing is designed to be deployed in a distributed manner where edge de-vices do the processing. In contrast, cloud computing is a more centralized concept.In Fog, processing and storage devices are located in close proximity compared to thecloud and this is the reason why Fog is more capable to serve latency aware servicesthrough access points, smart phones, base stations, switches, servers, and routers.The services which are referred to as low latency services are mostly emergency ser-vices including natural disasters, healthcare and so on. Apart from that, augmentedreality, video streaming, gaming and any other smart communication system alsorequires time- sensitive computation. With regards to improving quality of life viatechnology, Fog computing is going to play a major role in the near future. In thissection, we discuss several current research studies on Fog computing architecturebefore presenting a high-level Fog computing architecture.

1.2.1 Existing research on Fog computing architectureAs Fog computing has emerged recently, no standard architecture is available so farfor Fog computing paradigms [4]. Several studies [4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14,


3

15, 16] have proposed various architectures of Fog computing. The first Fog comput-ing architecture was depicted by Bonomi et al. [17] where the Fog layer was definedas distributed intelligence which resides in between the core network and sensor de-vices. Bonomi et al. [17] also point-out several characteristics which makes Fog anon-trivial extension of cloud computing. These characteristics are edge location,low latency, massive sensor network, very large number of nodes, mobility support,real-time interaction, dominant wireless connectivity, heterogeneity, interoperability,distributed deployment, on-line analytics and interplay with the cloud. Several types

Figure 1.1 Flow and processing of Big Data in IoT environment.


4

of architecture such as layer-based, hierarchical and network-based are proposed byvarious researchers for Fog computing as described below.

1.2.1.1 Fog layered architectureAazam et al. [4] presented a layered architecture of Fog where six different layerswere shown. Physical and virtual nodes as well as sensors were maintained and man-aged according to their service types and requirements by the lower layer known asthe physical and virtualization layer. The next upper layer is the Monitoring layerwhich monitors network and underlying node activities. This layer defines when andwhat task needs to be performed by which node and also takes care of the subsequenttask requirements that need be executed to complete the main task. The same layeralso monitors power consumption for energy constrained devices or nodes. Abovethe monitoring layer, the pre-processing layer resides which performs data manage-ment related tasks to get necessary and more meaningful data. After that, data isstored temporarily in the Fog resources by the next upper layer known as the tem-porary storage layer. Once the processed data is uploaded to the cloud, these areremoved from the local data storage media. For private data, the security layer pro-vides privacy, encryption and integrity measure related service. The topmost layeruploads pre-processed and secured data to the cloud. In this way, most processingwill be done in the Fog environment and allows the cloud to deal with more complexservices.

Arkian et al. [5] proposed a four layer Fog architecture: i) Data generator layer,ii) Cloud computing layer, iii) Fog computing layer and iv) Data consumer layer. Awide range of consumers is considered from individuals to enterprises in the dataconsumer layer. Consumers can submit their requests to three other layers and getresponses for the required services. The data generator layer is where the IoT de-vices reside and communicate with the cloud computing layer via the Fog comput-ing layer. In this architecture, data pre-processing will be done in the Fog computinglayer. This layer also enables context awareness and low latency. The cloud com-puting layer provides centralized control and a wide range of monitoring services.Long-term and complex behaviour analysis will be performed at this layer to supportdynamic decision making such as relationship modelling, long-term pattern recog-nition, and large-scale event detection. The key difference between this architecturewith others above is the direct communication between consumers and all three lay-ers.

The fog layer is presented as an intermediate layer between mobile devices andthe cloud in the Fog system architecture by Luan et al. [6]. According to this ar-chitecture, the main component of the Fog layer is the Fog server, which shouldbe deployed at a fixed location on the local premises of the mobile users. A Fogserver could be an existing network component like a base station or WiFi accesspoint. These servers communicate with mobile devices through single-hop wirelessconnections and provide them with pre-defined application services in its wirelesscoverage without seeking assistance from the cloud or other fog servers. This sys-tem architecture does not consider many other aspects but discusses the idea of a Fogserver.


5

Dastjerdi et al. [8] presented a five layer Fog computing reference architecture.The topmost layer is the IoT application layer which provides application facilitiesto the end users. The next lower layer is the software-defined resource managementlayer which deals with monitoring, provisioning, security, and management relatedissues. The next following layer is responsible for managing cloud services andresources. The next lower layer is the network layer which works to maintain theconnectivity between all devices and the cloud. The bottommost layer consists of enddevices such as sensors, edge devices, and gateways. It includes some apps whichbelong to this layer and these apps work by improving the device functionality. Inthis reference architecture, the Fog layer is completely absent and it also did nottestify where the computation is done.

A high-level architecture focusing on the communication in the Fog environ-ment is shown by Nadeem et al. [11] where communication among devices is shownin three different layers. A similar conceptual architecture was also presented byTaneja et al. [12] and Sarkar et al. [18] where the devices are located in three dif-ferent tiers. But Taneja et al. [12] and Sarkar et al. [18] added gateway devices toconnect the Fog devices and the cloud. These gateway devices are located in tier 2which is also denoted as the Fog layer.

1.2.1.2 Hierarchical Fog architectureGiang et al. [7] presented a hierarchical Fog architecture by classifying Fog devicesinto three different types based on their processing resources edge, computing, andinput/output nodes. Edge nodes execute actuation messages and generate sensingdata. Input-Output (IO) nodes have limited computing resources and maintain bro-kering communications with Edge nodes. Compute nodes have some computingresources to offer and this node is dynamic with a programmable runtime. Thesethree nodes can be implemented separately or with a combination based on systemdesigner preference.

A conceptual hierarchical architecture of Fog computing is presented by Hos-seinpour et al. [13] where the Fog computing layer is divided into three basic levelsand can be extended to N numbers of levels. Computation and storage are done atall levels except the lowermost level. Level 0 consists of sensors and actuators, level1 is named as a gateway Fog node and level 2 represents the core Fog nodes.

1.2.1.3 OpenFog architectureThe OpenFog architecture explanation is the most comprehensive one in which mostFog computing characteristics were considered [9]. However, OpenFog architecturedid not consider lower latency storage facilities near to business deployment and endusers. This architecture intends to do computation near to the end user to minimizelatency, migration costs and other network related costs along with bandwidth costs.Without synchronizing and routing all communication to the core network, low la-tency communication can take place and user requests are routed to the location clos-est to the end-users where computation elements are available. The implementationof management elements, including configuration and control management, and net-work measurements are deployed near to the endpoint rather than being controlled


6

from the gateway. In addition, the proposed architecture allows collection and pro-cessing of data using local analytics and the results are copied to the cloud in a securemanner for further processing and future use. Although they covered the maximumnumber of aspects about the Fog computing environment in their Fog computingarchitecture explanation, there is a lack of proper validation of their described archi-tecture through experimental deployment. However, they intend to collaborate withvarious related groups, including but not limited to, the Industrial Internet Consor-tium (IIC), ETSI-MEC (Mobile Edge Computing), Open Connectivity Foundation(OCF) and the Open Network function virtualization (OpenNFV).

1.2.1.4 Fog network architectureIntharawijitr et al. [10] illustrated the Fog network architecture and considered com-munication latency issues faced by Fog devices connected via a 5G cellular network.In their model, the edge router works as a Fog server and does the processing for theusers as mobile users send packets to the Fog server. The Fog server does not passthe request to the core network unless the request is for a cloud service. A mathe-matical model was defined to clarify communication delays and computing delays inthe Fog network and other related parameters. Three different policies were definedto choose a target Fog server for every task. To validate the proposed model, an ex-perimental evaluation was carried out in a simulation environment by employing theproposed policies.

1.2.1.5 Fog architecture for Internet of energyAn envisioned Fog of Everything (FoE) technology platform was proposed by Bac-carelli et al. [14]. In the architecture of this platform, all Fog devices (IoT sensors,smart car, smartphone or any other station) will be connected to the wireless base sta-tion via Fog to Things (F2T) and Things to Fog (T2F) two-way connectivity throughTCP/IP connections functioning onto IEEE802.11/15 single-hop links. All fog de-vices connected with the same base station are considered to be in the same cluster.All base stations are considered as Fog nodes and will be connected via Fog to Fog(F2F) links by the inter-Fog physical wireless backbone. Container-based virtualiza-tion is used to make a virtual clone associated with physical things. The virtualiza-tion layer supports efficient use of limited resources and generates the virtual cloneof physical things. The Fog node physical server serves cloned things and an overlayinter-clone virtual network was established which allows P2P communication amongclones by depending on TCP/IP end to end transport connections.

1.2.1.6 Fog computing Architecture based on nervous systemSun and Zhang [15] presented an architecture that was constructed based on thehuman nervous system. In their proposed architecture, the cloud data centre is con-sidered as the brain nerve centre, the Fog computing data centre is considered asthe spinal nerve centre and smart devices are considered as peripheral nerve centres.These three nerve centres spread their connections extensively throughout all of thebody of the system. Peripheral nerves scattered in the body and the brain will controlall the activities of the spinal cord. The structure of the system is designed based on


7

the neural structure of the human body where the brain is responsible in dealing withall tasks. All the smart devices connected to this system are referred to as the pe-ripheral nerves which are geographically distributed. These devices include tablets,phones, sensors, or smart watches. The Fog computing centre would address cer-tain simple and time-sensitive requests, for example, the spinal cord knee jerk reflexcan share the resource pressure of the cloud data centre. The spinal cord is the con-necting route between the brain and peripheral nerves; this is alike to the location ofthe fog data centre that joins the Internet of Things with high-level cloud data centres.

1.2.1.7 IFCIoT architectureIntegrated Fog cloud IoT (IFCIoT) architecture has been proposed by Munir et al.[16]. This architecture enables federated cloud services for IoT devices through anintermediary Fog infrastructure. The federated cloud is formed by multiple externaland internal cloud servers which match application and business needs. Gatewaydevices, smart routers, edge servers and base stations are the Fog nodes and muchof the processing takes place in these nodes. Fog nodes are autonomous; thus, each

Figure 1.2 Layered architecture of Fog Computing.


8

node can ensure uninterrupted service by their providers. The entire deployment ofthe fog computing environment can be local deployment in case of automation ofsingle office buildings and can also be distributed in the regional level, includinglocal levels in the case of large commercial companies located in multiple buildingsin various places in the IFCIoT architecture. This architecture supports distributeddeployment and information from various levels feed to a centralized system. Theconnectivity of all IoT devices is considered to be wireless connectivity via WLAN,WiMAX, and other cellular networks. Fog nodes maintain a connection with IoTdevices within its wireless range. The entire Fog is connected to the federated cloudservice through the core network. For collaborative processing, a Fog system couldbe connected to other Fogs wirelessly.

Table 1.1 summarizes the supported features of Fog computing architecturesproposed by various researchers in the past few years. According to this table, allproposed architectures represent the concept of data processing at the edge and usageof Fog devices for temporary storage whereas cloud infrastructure will be used forlong-term storage. Most of these architectures do not focus on virtualized devices asthe Fog device but rather represent the Fog device as just a physical device. How-ever, Fog devices can be virtualized (e.g. Cisco UCS server) and non-virtualized(e.g. smartphones). As reported by [9], scalability is the localized control, commandand processing; orchestration and analytics; and avoidance of network taxes. Au-tonomy represents flexibility, cognition, agility and value of data. Programmabilityis designated by programmable hardware and software, virtualization, multi-tenant,and app fluidity. However, these features were overlooked by most of the reviewedarchitectures. Some proposed architectures were also not validated by any model orexperimental environment [6, 9, 16]. Alternatively, many of them were validated bytheoretical models, simulations or experimental evaluations. Table 1.2 summarizesthe proposed models, performance metrics and simulation tools used by them.

1.2.2 High level Fog computing layered architectureAs shown in the layered architecture in Figure 1.2, end users are connected to theFog layer and this layer is responsible for maintaining communication with the datageneration and cloud computing layer. In Fog computing layered architecture, thereare four different layers: consumer layer, cloud computing layer, fog computinglayer and big data generation layer. This architecture is the high level architectureof Fog computing which separates the main actors of the Fog computing environ-ment. The consumer layer represents groups of people and institutes who will useFog computing services. Actual Fog computation processing will be done by thecloud and the Fog computing layer. Various groups of people or sectors can bene-fit from the Fog computing services. This includes individuals, health-care sectors,smart transportation systems, private organizations, government organizations, smartcities, academia and any other smart system that needs time-sensitive processing.The consumer groups mentioned above are connected with various sensors and datageneration devices in various ways. Based on the necessity of the particular group,they also have strict time and deadline constraint application requests. Healthcare


9

Tabl

e1.

1Fe

atur

esfo

cuse

dby

vari

ous

arch

itect

ures

.

Aut

hors

&Ye

ar

PhysicalFogDevices

VirtualizedFogdevices

FogServer

Monitoring

EnergyEfficiency

DataPreprocessing

TemporaryStorage

Security&Privacy

CloudStorage

Scalability

Autonomy

Programmability

ArchitectureValidation

Reliability

Aaz

amet

al.(

2015

)[4]

33

73

33

33

37

77

37

Lua

net

al.(

2015

)[6]

37

37

73

37

37

77

77

Gia

nget

al.(

2015

)[7]

37

77

73

37

33

73

37

Das

tjerd

ieta

l.(2

016)

[8]

37

73

73

33

33

77

37

Ope

nFog

(201

6)[9

]3

37

37

33

33

33

37

7

Ark

ian

etal

.(20

17)[

5]3

37

73

33

33

77

33

7

Bac

care

lliet

al.(

2017

)[14

]3

77

33

33

73

77

73

3

Sun

etal

.(20

17)[

15]

37

77

73

37

37

77

37

Mun

iret

al.(

2017

)[16

]3

33

33

33

73

37

77

7


10

Table 1.2 Models for Fog computing architecture evaluation.

Authors & Year Proposed Model SimulationTools Used

PerformanceMetrics

Aazam et al.(2015)[4]

Fog based IoTresourcemanagement

CloudSim 3.0.3

resourceprediction,resourceallocation, andpricing

Giang et al.(2015) [7]

DistributedDataflow (DDF)programming

N/A N/A

Dastjerdi et al.(2016) [8]

Dag of query forincidentdetection

CloudSimAverage tupledelay, corenetwork usage

Arkian et al.(2017) [5]

Cost-efficientresourceprovisioning

Not Mentioned

Service Latency,PowerConsumption,Cost

Baccarelli et al.(2017) [14] V-FoE testbed

iFogSim(Extension ofCloudSim)

EnergyConsumption,RTT, ConnectionFailure

Sun et al. (2017)[15]

Repeated gamebasedresource-sharingmodel

Not Mentioned SLA Violation,Completion Time

and transportation systems have the most sensitive applications in these cases. Al-though, the implementation of Fog computing will benefit everyone regardless ofgroup and organization as the cloud has been doing currently. The functionality ofthree main layers of Fog architecture is presented below.

1.2.2.1 Fog computing layerThe fog computing layer is the most important layer compared to the others since itmaintains communication with all other layers. The consumer layer directly sendsprocessing requests to this layer except communication with data generation and thecloud computing layer. This layer is responsible for data collection from the enddevices. This layer will decide whether or not a processing or storage request needsto be sent to the cloud. Various utilities for various services will be contained bythis layer. Small-scale data processing will be done in this layer with virtualizationsupport. The maximum processing is considered as stream processing and needsto be done in an online manner without storing huge amounts of data. However,


11

short-term storage and pre-processing will be done by this layer. All processingand storage related operations should be executed by any device that has processingpower and storage capacity. These devices are generally known as Fog devices andFog servers. It could be any type of device including routers, switches, access points,base stations, smartphone, servers, and hosts.

Application processing from the consumers can be done without the Fog layerwith help from the cloud. But, the problem is time-critical applications that cannotrely on the cloud because of the location of the cloud infrastructure. As an example:in a smart transportation system, we assume that if two alternate roads are availablefrom a specific point and one road becomes congested resulting in it taking sometime to update the status due to cloud processing, during this processing interval,some vehicles are directed to the congested road. As a consequence, the total traveltime to reach the destination is increased by half an hour. It may not affect ordi-nary people that much, but what if it happens to an ambulance directed to that route?It might even cost a life in some cases. To avoid such kinds of inconvenience, weshould not rely on the cloud to process sensor data. Other types of latency criticaldecisions are also possible in the smarter transportation system, such as dividing traf-fic into several routes during busy hours, natural disasters (rock falling on the road,extreme fog or rain etc.), as well as safety of pedestrians, animals, and cyclists. Theonly solution in these kinds of applications is to accomplish processing at the closestlocation near to the users. This is what the Fog layer actually does.

1.2.2.2 Data generation layerThe data generation layer contains all devices and sensors from where Big Data canbe generated. Big Data are generated in our everyday life. First of all, when weare at home, we are using various smart devices for communication, and many othersensors and actuators belong to lighting systems, music and entertainment systemsand alarm sensors are generating data by following certain intervals. Secondly, ourregular activities such as shopping, working out, office work and school work lead usto generate Big Data. Finally, our regular hits to various services also produces datathat might have useful insight, for example, we need to know traffic updates, tasklists, offers in the various shops, activities, and tasks of the school. In this way, datais generated at every moment. We also find many smart devices and sensors every-where. Security cameras, roadside dash cams, speed cameras, temperature sensors,GPS sensors, alarm sensors and actuators are generating data every jiff as we keepmoving. We are also using the smart device for tracking our exercise and other ac-tivities or even sleeping activities. Hence, all devices that can generate data belongto this layer and transfer generated data to the Fog computing layer. Any action-able request which is generally done by the actuator will be initiated from the Fogcomputing layer when necessary. Thus, this layer has two-way communication withthe Fog computing layer. Data generated by the Big Data generation layer will becollected and analysed by the Fog computing layer. The Fog computing layer willdecide whether storage and processing will be done by its own resources or if there


12

is a need to use cloud resources.

1.2.2.3 Cloud computing layerTraditional cloud infrastructure resides in this layer. The main functionality of thislayer is to provide long-term storage and complex time insensitive processing. Asthis layer cannot communicate directly to the consumers and end devices, it com-pletely depends on the Fog computing layer for communication. From the usersperspective, users are no longer submitting their request to the cloud; they are sub-mitting requests to the Fog computing layer. As a result, processing will be donemore quickly when users are depending on Fog computing resources which are lo-cated in the close proximity. It is not possible to eliminate the cloud computing layerbecause it assumes that the devices located in the Fog computing layer have minimalcomputing and storage resources. Thus, if we need to store a high volume of dataand process the same, we would have no other option but to depend on the cloud.

1.3 Limitation of the cloud to execute Big Data applications

1.3.1 Exploding generation of sensor dataDue to the high pace of IoT technology adoption, Big Data generation is increasingexcessively. It is expected that the total number of connected devices worldwide willbe about 30 billion by 2020 and it will further increase to 80 billion by 2025 whichis nearly triple within a five-year gap [19]. It is also predicted that by 2025, about152,000 new devices will be connected to the Internet every minute [19]. Accordingto an IDC report, by 2020, about 40% of data will be processed at the edge [20]. It ishard to send all generated data into the cloud due to communication costs. Hence, itis better to process the generated data near to the users without sending to the cloud.IoT has been crafted to create greater opportunities that can help reshape the modernworld via increases in revenue and reductions in operational costs by revealing betterinsight where collections of huge data alone are not sufficient. To get the real benefitof IoT revolution, organizations need to develop a platform where it is possible togather, manage and evaluate massive sensor-generated data in an efficient and cost-effective manner [5].

1.3.2 Inefficient use of Network bandwidthThe cloud is the best place to process Big Data because it has high computation andprocessing resources. But sending all generated data to the cloud for processing willsignificantly increase network traffic. Hence, it is necessary to reduce the volumeof data at the edge as well as to mine data to find the pattern at the edge level.Some applications, video surveillance systems for example, generate a high volumeof video data. There are two critical issues arising from such systems. Firstly, if wesend all video data to the cloud, it may occupy the maximum available bandwidthbefore sending them to the cloud. Secondly, processing these data at the edge is alsochallenging because it needs a large amount of processing power. As we mentioned


13

earlier, the cloud is the best for the processing of high volumes of data, but theproblem is that we do not have unlimited available bandwidth. The complexity ofsuch systems will be increased day by day while video data are important for crimecontrol and the implementation of a smart monitoring system in a smart city. So,the cloud is not the convenient way to deal with the applications that need to processvideo files.

1.3.3 Latency awarenessApplications related to augmented reality, online gaming, smart homes and smarttraffic are more latency sensitive. Fog nodes are usually located one or two hopdistances from the user. Hence, if the Fog concept is used for Big Data processingat the edge, it can easily support latency aware applications. In contrast, the cloudis located in a multiple hop distance which is quite far from the user where latencyis higher than Fog. Latency also affects online businesses. As examples, a 100msdelay may cause a 1% reduction in Amazon sales and a 500ms delay would cause a20% drop in Google traffic [21].

1.3.4 Location awarenessMost IoT applications are context aware, meaning that application processing de-pends on the location and other applications running nearby. Sending all thesecontext-aware application requests to the cloud is not realistic and sometimes notaffordable due to bandwidth and delay constraints. Many IoT applications also re-quire minimum processing delays of less than a number of milliseconds [22] such ashealth-care applications, vehicular networks, drone control applications and emer-gency response systems. Such applications also require real time data processing.These types of application services always depend on the surrounding environmen-tal data rather than information available to other locations.

Location awareness is also used in smart traffic applications to detect the patternof the traffic such as roadwork, roadblocks, traffic congestion and accidents. Theseapplications share information among connected vehicles to improve vehicle naviga-tion and traffic management [23]. Another example is the smart surveillance appli-cation system where a police officer of a local police station can see the video streamof suspicious people around his designated area and is able to track and initializeaction to prevent damage to the public [24]. In the above scenarios, processing andrunning applications on the cloud may not efficient enough due to the high responsetimes of the cloud.

1.4 Challenges faced when executing Big Data applications onFog

Data pre-processing and post-processing is completely dependent on the cloud whilewe depend on the cloud for Big Data processing. On the other hand, in Fog, BigData streams are pre-processed locally by using a number of spatially distributed


14

stationary or mobile devices. Then, the processed data are sent to the cloud datacentre for further post-processing while necessary. Pre-processing, transformation,and post-processing can be referred to as a three-phase life cycle of data processingin Fog. Fog devices can be mobile and they also have resource-limited properties,therefore, there are many obstacles that exist when executing Big Data applicationson Fog. This section points out the key issues of Big Data application execution onFog.

1.4.1 Resource limited Fog deviceFog devices used for computation have limited resources for processing. Comparedto the cloud, the resources of Fog devices are very limited. However, Big Data canbe executed on these devices by making Fog clusters as Google uses commodity ma-chines as the server and also introduced the map-reduce concept to deal with hugedata processing using commodity machines. Any device that has computation capa-bilities can act as a Fog device, from user devices to all kinds of network managementdevices are considered as Fog devices. Generally, these devices have their own op-erating systems and applications that occupy most of the system resources. RunningFog applications on these devices is always the second priority. Thus, efficient andintelligent job allocation policy is required in order to protect a high number of jobfailures.

1.4.2 Power LimitationFirst of all, in the Fog paradigm, any device that has processing, storage and networkcapabilities can act as a Fog device. Hence, battery-powered mobile devices such aslaptops, smartphones and tablets can act as a Fog device [25]. The energy of thesedevices is limited and due to this, the device may turn off from low battery levelsduring computation. As a consequence, tasks that were running before power offneed to be rescheduled to another device. Thus, the scheduler might aware of poweravailability before task allocation to a mobile device. Secondly, many types of sensordevices are powered by batteries and these batteries should be recharged frequently.It is a challenging issue in some cases as failure due to the battery may cause seriousharm. For example, body sensor networks which are monitoring patients. In suchscenarios, we should use energy efficient methods by incorporating primary dataanalytics which will reduce the volume of data that needs to be stored and transmitted[26].

1.4.3 Selection of Master nodeIn the Fog environment, many of the devices might be mobile and these devices arealways moving which may cause a problem to make the cluster of Fog devices. BigData processing using map-reduce where it is necessary to select a master node ismainly responsible for tracking jobs. If the master node moved to another cluster,then it will be necessary to execute the whole job again. To address this issue of


15

master node selection in the Fog cluster is challenging. However, selecting a masternode among stationary nodes is a good solution.

1.4.4 ConnectivityMost of the Fog and IoT devices are connected using wireless connectivity whichhas problems with interference and fading. These problems cause fluctuations ofthe access bandwidth. Beside this, in the Fog environment thousands of devices andsensors coexist and communicate with each other repeatedly. Also, the number ofconnected devices will be increasing over time. To address this issue, it is crucial todevelop an efficient model that can estimate the number of connected devices withina Fog device and can predict resource incorporation before the failure of the system.

1.5 Recent advances on Big Data application execution on Fog

Fog computing is a comparatively recent research trend and much research workhas been done in this area. However, only a few research works have addressedspecifically Big Data application execution on the Fog paradigm. Most of the workhas been focused on the healthcare system and also some work has been done invarious areas including smart cities and virtual learning systems.

Dubey et al. [26] proposed a service-oriented architecture for telehealth BigData in Fog computing by adding low powered embedded computers in the Foglayer which performs data analytics and data mining on raw data. They consideredtwo use cases, speech motor disorder and cardiovascular detection. Raw data forboth use cases were processed in the Fog layer. After the processing of raw datafrom cardiovascular detection sensors or speech motor disorder sensors, all detectedpatterns are stored and the unique, distinctive pattern is sent to the cloud for furtherprocessing. In the first case study, speech data was sent to the Fog device by thesmartwatch. Then, the Fog device does three steps of processing, feature extrac-tion, pattern mining and compression. In this way, it converts speech into averagefundamental frequency and loudness features. The compressed speech data is thensent to the cloud. The cloud processes the received average fundamental frequencyand loudness to convert it into original speech time-series. For this case study, Dy-namic Time Warping (DTW) was used for speech data mining and Clinical SpeechProcessing Chain (CLIP) algorithm was used for computing relevant clinical param-eters such as fundamental frequency and loudness. In the other case study on ECGmonitoring, DTW was used for pattern recognition, and after recognizing the pattern,the processed data was sent to the cloud for further processing.

Ahmad et al. [27] proposed a framework health monitoring in the Fog paradigmwhere the Fog is an intermediate layer between cloud and users. The proposed de-sign feature reduced communication costs compared to other available systems. Thework also proposed utilizing the Cloud Access Security Broker (CASB) to deal withsecurity and privacy issues. The sensory data will be generated by a smartphone in3 second intervals and will be sent to a local machine known as the Health Fog inevery minute interval in a batch. Besides activity data, other sensory data from smart


16

homes and hospital activities will be stored in the Health Fog. Then, the intermediateprocessing will be done on Fog and shared with nutritionists or doctors as per userpreference. The final processed information on calorie burning and activity detectionwill be deposited in the cloud and shared according to user preference.

Tang et al. [28, 29] studied smart pipeline monitoring in smart cities to findthreatening events by using a sequential learning algorithm which is based on usingfibre optic sensors to analyse Big Data in Fog computing. The authors have pre-sented 4-Layer architecture for monitoring the pipeline smartly and it can be usedfor other infrastructures like smart buildings and smart traffic. The lowest layer oftheir architecture is the fibre optic sensor network where the cross-correlation methodand time-domain filter is used to detect changes in the physical environment such asstress, temperature and strain. The next upper layer consists of small computingnodes working in parallel. Each node is responsible to perform two tasks, the first isto detect potential threat patterns from sensor data using machine learning algorithmsand the second is the feature extraction to do the further processing by the upper layerknown as intermediate computing nodes. To reduce communication load, raw datafrom sensors will not be received by the intermediate computing nodes. This layercoordinates data analysis from various locations to identify hazardous events in aspecific region. The most upper layer is the cloud which is basically built by usingHadoop clusters to determine and predict long-term natural disasters. The hiddenMarkov model is used as a sequential learning method and has been verified suc-cessfully in a real environment.

Zhang et al. [30] presented a hierarchical model for resource management ininter and intra Fog environment which considered packet loss and energy efficiencyof the intelligent vehicle and Fog server. The model is comprised of two layers:Fog layer and edge layer. The Fog layer is the association of local Fog servers,clouds, and coordination servers. On the other hand, Vehicular Ad-hoc Networks(VANETs), cellular networks and IoT applications are the key elements of the edgelayer. The Intra Fog resource management manages internal tasks assigned to thevirtual machine within the Fog server and the task size is ordered by an AdaptiveLoad Dispatcher (ALD) where the virtual machine processing rate is adjustable. Ininter Fog resource management operations, all local Fog servers update their workingstates to a server, which coordinates all Fog servers and assigns overflow workloadto idle Fog servers nearby. This inter-fog resource management operation controlsthe data flow with the help of access control routers without disturbing intra-fogoperations.

A distributed resource sharing scheme is proposed by Yin et al. [31] wherethe Software Defined Network (SDN) controller dynamically adjusts the quantity ofapplication data that will be pre-processed by the Fog nodes for Big Data stream-ing applications. The problem of application data quantity assignment is formulatedas a social welfare maximization problem in their work. Also, the loss of infor-mation value occurring in the pre-processing process was contemplated. A HybridAlternating Direction Method of Multipliers (H-ADMM) algorithm was employedto solve computation burdens in a fully distributed environment composed of Fognodes, SDN controllers and cloud. Moreover, an efficient message exchange pat-


17

tern was integrated to reduce communication costs on the SDN controller when itsdealing with a large number of Fog nodes.

Pecori [32] proposed a virtual learning architecture where Big Data streamswere processed in a Fog computing environment using Apache Storm and Scal-able Advanced Massive Online Analysis (SAMOA) as distributed machine learningframework. The Storm real-time computation system was chosen for implementa-tion due to its inbuilt master-slave architecture that can be mirrored easily betweenFog and cloud. The reason behind choosing Storm instead of the spark is that theStorm is event-oriented rather than batching data updates at a routine interval. TheSAMOA was selected for the deployment of its finer integration with Storm. In theproposed architecture, cloud infrastructure was used for high volume storage, histor-ical backup operation and mining jobs with high latency. The cloud also provideslong-term forecasts and scores of selected features by communicating with inter-mediate and macro users such as educational institution managers and the policymakers through outer APIs. The Fog layer is composed of lightweight storage facili-ties with distributed NoSQL support along with multiple Storm slaves. These slavescan be network gateways, sensors or any other smart devices and they can performshort-term predictions by using light mining techniques. These predictions provideuseful suggestions to all consumers, students, tutors and teachers in an e-learningenvironment. Using the above proposed technique, the Fog based stream processingvirtual learning framework would be beneficial for students and instructors as wellas distance learning institutions.

A summary of the above Fog based Big Data application frameworks and archi-tectures are presented in Table 1.3 and Table 1.4 with their limitations.

1.6 Fog computing products

1.6.1 Cisco IOxThe Cisco IOx application platform enables Fog processing near to the end devicesby integrating IoT sensors and the cloud. Cisco IOx is the combination of four com-ponents: IOS and Linux OS, Fog Director, the SDK and development tools, and Fogapplications. Cisco IOS is the leading NOS which ensures secure network connec-tivity and Linux is an open source customizable platform. However, this platform isdesigned to execute applications on the Cisco IoT network infrastructure. Fog Di-rector helps to administrate applications over the network running on the Cisco IOxenvironment. The SDK and development tools provide a collection of tools, com-mand line utility and web-based applications for the developers. Fog applicationsare readily executable applications that will run on IOx enabled infrastructure [33].Cisco 800 series routers support IOx and it can be used as a Fog device because itsupports two OS on two cores. One core dedicated to IOS and the other core supportsLinux based OS which helps to do the processing in the fog environment [34].


18

Tabl

e1.

3R

esea

rch

wor

ksth

atfo

cuse

don

both

Fog

and

Big

Dat

a.

Aut

hors

&Ye

arA

ddre

ssB

igD

ata

App

licat

ions

Fog

Com

-pu

ting

Para

digm

App

licat

ion

Type

Lay

ersi

nA

rchi

tec-

ture

Dis

trib

uted

Proc

essi

ngFr

amew

ork

for

Clo

ud

Dis

trib

uted

Proc

essi

ngFr

amew

ork

for

Fog

Dev

ices

used

asFo

g

Tang

etal

.(20

15,

2017

)[29

,28]

Yes

Yes

Smar

tCiti

es4-

Lay

ers

Had

oop

Not

men

tione

dN

otm

entio

ned

Dub

eyet

al.

(201

5)[2

6]Y

esY

esTe

lehe

alth

3-L

ayer

sN

otm

entio

ned

Not

men

tione

dIn

telE

diso

n

Ahm

adet

al.

(201

6)[2

7]Y

esY

esH

ealth

care

3-L

ayer

sN

otm

entio

ned

Not

men

tione

d

Win

dow

sba

sed

mac

hine

Zha

nget

al.

(201

7)[3

0]Y

esY

esSm

artc

ities

2-L

ayer

sN

AN

AN

A

Yin

etal

.(2

017)

[31]

Yes

Yes

Dat

aas

sign

men

tto

each

Fog

node

3-L

ayer

sN

AN

AN

A

Peco

ri(2

018)

[32]

Yes

Yes

Vis

ual

lear

ning

3-L

ayer

sSt

orm

Stor

mSm

artp

hone


19

Tabl

e1.

4A

lgor

ithm

s,ar

chite

ctur

e,fr

amew

ork

and

limita

tion

ofva

riou

sre

sear

chw

orks

onFo

gco

mpu

ting

for

Big

Dat

a.

Aut

hors

&Ye

arA

lgor

ithm

sUse

dPr

opos

edA

rchi

tect

ure

orFr

amew

ork

Lim

itatio

n

Tang

etal

.(20

15,

2017

)[29

,28]

Sequ

entia

llea

rnin

gal

gori

thm

s

Big

Dat

aan

alys

isar

chite

ctur

ein

Fog

com

putin

g

Fixe

dsp

eed

ofco

mpu

ting

node

san

ddi

dno

tcon

side

rm

emor

yac

cess

time

Dub

eyet

al.(

2015

)[26

]D

ynam

icTi

me

War

ping

(DT

W)a

ndC

linic

alSp

eech

Proc

essi

ngC

hain

(CL

IP)

Serv

ice

orie

nted

arch

itect

ure

forF

ogco

mpu

ting

Com

plex

spee

chan

alys

isto

reco

gniz

em

ore

accu

rate

spee

chdi

sord

er

Ahm

adet

al.(

2016

)[27

]H

omom

orph

icen

cryp

tion

Fram

ewor

kfo

rhea

lthan

dw

elln

ess

appl

icat

ions

Dis

trib

uted

Fog

envi

ronm

enti

sno

tpre

sent

Zha

nget

al.(

2017

)[30

]N

AH

iera

rchi

calr

esou

rce

man

agem

entm

odel

fori

ntra

and

inte

rFog

Ado

ptio

nof

the

mod

elin

real

envi

ronm

ent

Yin

etal

.(20

17)[

31]

Hyb

rid

Alte

rnat

ing

Dir

ectio

nM

etho

dof

Mul

tiplie

rs(H

-AD

MM

)al

gori

thm

Dis

trib

uted

reso

urce

shar

ing

sche

me

with

SDN

cont

rolle

rD

oes

notc

onsi

deri

ncen

tive

mec

hani

sm

Peco

ri(2

018)

[32]

Dis

trib

uted

Has

hTa

bles

(DH

T)a

ndM

achi

nele

arni

ngal

gori

thm

s

e-le

arni

ngar

chite

ctur

ew

ithth

ein

tegr

atio

nof

Clo

ud

Fog

and

Big

Dat

aO

verl

ooke

dth

ese

curi

tyan

dpr

ivac

yco

ncer

n


20

1.6.2 LocalGrid’s Fog computing platformThis is an embedded software that can be installed on network edge devices. Itenables various communication protocols into an open standard by the LocalGridFog computing platform. The embedded software can be installed on all sensors,switches, routers, machines and other edge devices which allow secure access forFog processing and also leverage the communication gaps between new and legacydevices. It also forms a P2P communication between multiple edge devices and facil-itates real-time coordination and control without a centralized server. The LocalGridalso has a communication infrastructure with the cloud by incorporating intelligencenetwork communication which reduces latency, improves security and reduces us-age of bandwidth. On this platform, data processing and decision making on edgedevices can be done on a distributed way [35].

1.6.3 Fog Device and GatewaysBesides proprietary products, there are many other alternatives that can be used forFog devices and gateways commonly known as a computer on modules. These de-vices include Intel Edison, Raspberry Pi, Arduino, Asus Tinker Board, Odroid-C2,Banana Pi M2 Ultra, and Odroid-XU4. Among them, the Raspberry Pi is most pop-ular. However, we may find many other vendors whose are producing more powerfuldevices than the Raspberry Pi ata competitive price. Most of these devices supportAndroid OS and can use other types of OS, including Raspbian, OSMC, OpenELEC,Windows IoT Core, RISC OS and Ubuntu MATE. Many research studies have beendone using these computers on modules [26, 36, 37, 38, 39, 40] in the past couple ofyears to implement the Fog environment for Big Data processing.

1.7 Research Issues

There are several research issues currently available that need to be addressed forBig Data application execution on Fog. Figure 1.3 represents several research issuesin this area.

To run Big Data IoT applications in Fog, stream processing is done in the Fogenvironment and Big Data processing will be done in the cloud. However, it is pos-sible to implement Fog cluster near to users using Fog devices and have potential toexecute Big Data related processing to some extent. To do so, one of the key issues iscross-layer communication between the Fog to the cloud and the end user to the Fog.Besides cross-layer communication, inter Fog communication is also equally impor-tant. In inter Fog communication, data transmission is contested by heterogeneousservice policies, network topology and device connections [6]. Moreover, integrationof emerging technologies such as 5G technologies, Software Defined Networking(SDN) and Network Function Virtualization (NFV) is also necessary to incorporatethese technologies with Fog.

With regards to the deployment, the main challenges are application placementsand scaling. First of all, network operators need to customize the applications basedon local demand. Secondly, due resource-limited properties of Fog devices, it is


21

Figure 1.3 Research issues of Fog computing for Big Data application execution.

difficult to scale resources as per user demand. Thirdly, since the location is moreimportant in Fog environment, placing Fog servers in the right place is also challeng-ing [6].

To satisfy low-latency requirements, pre-cache technology needs to be employedand Fog nodes should store cache to the Geo-distributed nodes by passively predict-ing users demands. In this way, delays when downloading content will be reducedsignificantly and efficient use of storage resources fulfils Fog application require-ments. The storage resource of Fog device is limited so storage expansion technolo-gies are very effective in order to improve overall service quality [41].

Resource management in Fog has the utmost priority to maintain the lifespanand performance. Hence, resource scheduling techniques including placement, con-solidation, and migration need to be investigated extensively [41]. Application place-ment to these resources is another research challenge for time-critical computing ap-plications such as emergency response, healthcare, vehicular network, augmentedreality, online gaming, brain-machine interface and any other smart environment re-lated applications. Besides the above challenges, energy management, programmingplatforms, security, privacy, and scalability are also important research issues. Alsouser’s privacy [42] is most important that need to be explore in Fog perspective.


22

1.8 Conclusion

Fog computing is an emerging technology which has flourished in solving Big DataIoT application execution problems by processing continuously generated data at theedge. This computing paradigm is a high-potential computing model that is grow-ing rapidly but it is not mature enough as many issues still need to be investigatedextensively. This paper reviewed and presented several existing architecture of Fogcomputing in order to identify the research issues related to Big Data applicationexecution using Fog paradigm. We presented a high-level Fog computing architec-ture and discussed many other architectures of Fog computing and highlighted thefeatures of numerous proposed architectures. We also discussed key limitations ofthe cloud to execute Big Data applications, especially in the IoT environment. Fol-lowing the limitations of cloud, some challenges to execute Big Data application onFog were presented. Also, some recent research works that specifically addressedBig Data application executions on Fog were investigated. Consequently, the char-acteristics of some currently available commercial Fog related platforms and deviceswere discussed. Finally, several open research issues were presented. Hopefully,these will pave future research directions among industry experts and academia.


References

[1] Vermesan O, Friess P. Internet of things-from research and innovation tomarket deployment. vol. 29. River Publishers Aalborg; 2014. p. 7–112.

[2] Tammishetty S, Ragunathan T, Battula SK, et al. IOT-Based Traffic SignalControl Technique for Helping Emergency Vehicles. In: Proceedings of theFirst International Conference on Computational Intelligence and Informat-ics. Springer; 2017. p. 433–440.

[3] Yousefpour A, Ishigaki G, Gour R, et al. On Reducing IoT Service Delay viaFog Offloading. IEEE Internet of Things Journal. 2018;PP:1–12.

[4] Aazam M, Huh EN. Fog computing micro datacenter based dynamic resourceestimation and pricing model for IoT. In: Proceeding of the 29th IEEE Inter-national Conference on Advanced Information Networking and Applications(AINA). IEEE; 2015. p. 687–694.

[5] Arkian HR, Diyanat A, Pourkhalili A. MIST: Fog-based data analyticsscheme with cost-efficient resource provisioning for IoT crowdsensing appli-cations. Journal of Network and Computer Applications. 2017;82:152–165.

[6] Luan TH, Gao L, Li Z, et al. Fog computing: Focusing on mobile users atthe edge. Networking and Internet Architecture. 2015;Available from: https://arxiv.org/abs/1502.01815.

[7] Giang NK, Blackstock M, Lea R, et al. Developing iot applications in thefog: A distributed dataflow approach. In: Proceeding of the 5th InternationalConference on the Internet of Things (IOT). IEEE; 2015. p. 155–162.

[8] Dastjerdi AV, Gupta H, Calheiros RN, et al. Fog computing: Principles,architectures, and applications. In: Kaufmann M, editor. Internet of Things:Principle & Paradigms. USA; 2016. p. 1–26.

[9] Group OCAW, et al. OpenFog Architecture Overview. White Paper, Febru-ary. 2016;p. 1–35.

[10] Intharawijitr K, Iida K, Koga H. Analysis of fog model considering comput-ing and communication latency in 5G cellular networks. In: Pervasive Com-puting and Communication Workshops (PerCom Workshops), 2016 IEEE In-ternational Conference on. IEEE; 2016. p. 1–4.

[11] Nadeem MA, Saeed MA. Fog computing: An emerging paradigm. In: Pro-ceeding of the 6th International Conference on Innovative Computing Tech-nology (INTECH). IEEE; 2016. p. 83–86.

[12] Taneja M, Davy A. Resource Aware Placement of Data Analytics Platformin Fog Computing. Procedia Computer Science. 2016;97:153–156.

[13] Hosseinpour F, Plosila J, Tenhunen H. An Approach for Smart Managementof Big Data in the Fog Computing Context. In: Proceeding of the IEEE Inter-

https://arxiv.org/abs/1502.01815

https://arxiv.org/abs/1502.01815


24 REFERENCES

national Conference on Cloud Computing Technology and Science (Cloud-Com). IEEE; 2016. p. 468–471.

[14] Baccarelli E, Naranjo PGV, Scarpiniti M, et al. Fog of Everything: Energy-efficient Networked Computing Architectures, Research Challenges, and aCase Study. IEEE Access. 2017;5:9882–9910.

[15] Sun Y, Zhang N. A resource-sharing model based on a repeated game in fogcomputing. Saudi journal of biological sciences. 2017;24(3):687–694.

[16] Munir A, Kansakar P, Khan SU. IFCIoT: Integrated Fog Cloud IoT: A novelarchitectural paradigm for the future Internet of Things. IEEE ConsumerElectronics Magazine. 2017;6(3):74–82.

[17] Bonomi F, Milito R, Zhu J, et al. Fog computing and its role in the internet ofthings. In: Proceedings of the first edition of the MCC workshop on Mobilecloud computing. ACM; 2012. p. 13–16.

[18] Sarkar S, Misra S. Theoretical modelling of fog computing: a green comput-ing paradigm to support IoT applications. IET Networks. 2016;5(2):23–29.

[19] Kanellos M. 152,000 Smart Devices Every Minute In 2025: IDC OutlinesThe Future of Smart Things. Forbes Magazine; 2016. Available from:https://www.forbes.com/sites/michaelkanellos/2016/03/03/152000-smart-devices-every-minute-in-2025-idc-outlines-the-future-of-smart-things/#3c77fde84b63.

[20] Connecting the IoT: the road to success;. Available from: https://www.idc.com/infographics/IoT.

[21] Greenberg AG, et al. Networking the Cloud. In: ICDCS; 2009. p. 264.[22] Zhang Y, Ren J, Liu J, et al. A survey on emerging computing paradigms for

big data. Chinese Journal of Electronics. 2017;26(1):1–12.[23] Koldehofe B, Ottenwalder B, Rothermel K, et al. Moving range queries in

distributed complex event processing. In: Proceedings of the 6th ACM In-ternational Conference on Distributed Event-Based Systems. ACM; 2012. p.201–212.

[24] Hong K, Smaldone S, Shin J, et al. Target container: A target-centric par-allel programming abstraction for video-based surveillance. In: DistributedSmart Cameras (ICDSC), 2011 Fifth ACM/IEEE International Conferenceon. IEEE; 2011. p. 1–8.

[25] Naha RK, Garg S, Georgakopoulos D, et al. Fog Computing: Survey ofTrends, Architectures, Requirements, and Research Directions. IEEE access.2018;6:47980–48009.

[26] Dubey H, Yang J, Constant N, et al. Fog data: Enhancing telehealth big datathrough fog computing. In: Proceedings of the ASE BigData & SocialInfor-matics 2015. ACM; 2015. p. 14.

[27] Ahmad M, Bilal M, Hussain S, et al. Health Fog : a novel frameworkfor health and wellness applications. The Journal of Supercomputing.2016;72(10):3677–3695.

[28] Tang B, Chen Z, Hefferman G, et al. Incorporating Intelligence in Fog Com-puting for Big Data Analysis in Smart Cities. IEEE Transactions on IndustrialInformatics. 2017;13(5):2140–2150.

https://www.forbes.com/sites/michaelkanellos/2016/03/03/152000-smart-devices-every-minute-in-2025-idc-outlines-the-future-of-smart-things/#3c77fde84b63



https://www.idc.com/infographics/IoT

https://www.idc.com/infographics/IoT


REFERENCES 25

[29] Tang B, Chen Z, Hefferman G, et al. A hierarchical distributed fog computingarchitecture for big data analysis in smart cities. In: Proceedings of the ASEBigData & SocialInformatics 2015. ACM; 2015. p. 28.

[30] Zhang W, Zhang Z, Chao Hc. Cooperative Fog Computing for Dealing withBig Data in the Internet of Vehicles : Architecture and Hierarchical ResourceManagement. 2017;(December):60–67.

[31] Yin B, Shen W, Cheng Y, et al. Distributed Resource Sharing in Fog-assistedBig Data Streaming. 2017;p. 1–6.

[32] Pecori R. A Virtual Learning Architecture Enhanced by Fog Computing andBig Data Streams. 2018;p. 1–30.

[33] Cisco IOx; 2016. Available from: https://www.cisco.com/c/en/us/products/cloud-systems-management/iox/index.html.

[34] Cisco IOx Local Manager Pages and Options; 2016. Available from:http://www.cisco.com/c/en/us/td/docs/routers/access/800/software/guides/iox/lm/reference-guide/1-0/iox local manager ref guide/ui reference.html.

[35] LocalGrid Fog computing platform datasheet; 2015. Available from:http://www.localgridtech.com/wp-content/uploads/2015/02/LocalGrid-Fog-Computing-Platform-Datasheet.pdf.

[36] Barik R, Dubey H, Sasane S, et al. Fog2Fog: Augmenting Scalability in FogComputing for Health GIS Systems. In: Connected Health: Applications,Systems and Engineering Technologies (CHASE), 2017 IEEE/ACM Interna-tional Conference on. IEEE; 2017. p. 241–242.

[37] Barik RK, Dubey AC, Tripathi A, et al. Mist Data: Leveraging Mist Com-puting for Secure and Scalable Architecture for Smart and Connected Health.Procedia Computer Science. 2018;125:647–653.

[38] Hajji W, Tso FP. Understanding the performance of low power Raspberry PiCloud for big data. Electronics. 2016;5(2):29.

[39] He J, Wei J, Chen K, et al. Multi-tier fog computing with large-scale IoT dataanalytics for smart cities. IEEE Internet of Things Journal. 2017;PP(99):1–10.

[40] Giordano A, Spezzano G, Vinci A. Smart agents and fog computing for smartcity applications. In: International Conference on Smart Cities. Springer;2016. p. 137–146.

[41] Hu P, Dhelim S, Ning H, et al. Survey on fog computing: architecture, keytechnologies, applications and open issues. Journal of Network and ComputerApplications. 2017;98:27–42.

[42] Aghasian E, Garg S, Montgomery J. User’s Privacy in RecommendationSystems Applying Online Social Network Data, A Survey and Taxonomy.arXiv preprint arXiv:180607629. 2018;.

https://www.cisco.com/c/en/us/products/cloud-systems-management/iox/index.html

https://www.cisco.com/c/en/us/products/cloud-systems-management/iox/index.html

http://www.cisco.com/c/en/us/td/docs/routers/access/800/software/guides/iox/lm/reference-guide/1-0/iox_local_manager_ref_guide/ui_reference.html

http://www.cisco.com/c/en/us/td/docs/routers/access/800/software/guides/iox/lm/reference-guide/1-0/iox_local_manager_ref_guide/ui_reference.html

http://www.localgridtech.com/wp-content/uploads/2015/02/LocalGrid-Fog-Computing-Platform-Datasheet.pdf

http://www.localgridtech.com/wp-content/uploads/2015/02/LocalGrid-Fog-Computing-Platform-Datasheet.pdf

Documents

arXiv:1811.09047v1 [cs.DC] 22 Nov 2018 · smarter planet [1, 2]. From the IoT environment, Big Data are generated in each and every moment from sensors, messaging systems, mobile